When: Monday, March 8th, 2021

Variational Inference for Cataloging the Visible Universe

A key task in astronomy is to locate astronomical objects in images and to characterize them according to physical parameters such as color, apparent magnitude, and morphology. This task, known as cataloging, is challenging for several reasons: many astronomical objects are much dimmer than the sky background, labeled data are generally unavailable, overlapping astronomical objects must be resolved collectively, galaxy shapes are complex, and the datasets are enormous. In this talk, I present a new approach to cataloging based on inference in a fully specified probabilistic model. To approximate the posterior distribution, I propose a procedure based on variational inference that employs a deep neural network to “amortize” the computational cost over regions of sky. For crowded starfields, the proposed method is as much as 10,000 times faster than existing methods, while producing more accurate results. Preliminary results indicate that the proposed method is also effective at cataloging heavily blended galaxies.